Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
While advances in hardware are making it practical to build large-scale sensor networks, methods for efficiently extracting information from them remain an open area of research. As sensor networks are scaled-up, resource limitations become a key issue. This leads to an inherent need for approximate, satisficing problem solving in large-scale sensor networks. The PIs believe that designs based on systems of intelligent agents offer many potential advantages. Distributed sensor interpretation (DSI) has been the subject of considerable research within the multi-agent systems (MAS) community, and the PIs have been heavily involved in this research.
Unfortunately, the state-of-the-art in MAS is not yet such that there are formal design methodologies that allow a potential DSI domain to be analyzed and an optimal system design determined. %This is a serious issue hindering the widespread adoption of MAS-based approaches. The PIs have been engaged in several lines of research---largely sponsored by NSF---to develop methods that can improve the process of designing MAS-based DSI systems. As part of this work, they have been investigating the use of decision theoretic (DT) approaches to build DSI systems. There are two main components of such an approach: (1) distributed Bayesian networks (DBNs) and (2) decentralized Markov decision processes (DEC-MDPs). The PIs have used these techniques in very small DSI systems to develop optimal coordination strategies to achieve a specified level of solution quality. What is exciting about this work is that it demonstrates how a formal, principled approach can be used to design satisficing MAS strategies. The PIs have come to believe that DT approaches represent one of the most promising directions for future MAS research on DSI and many other applications.
The basic goal of the research being proposed here is to extend the decision theoretic approaches currently being studied by the PIs, so they can be applied to larger-scale, real-world DSI problems. The main objectives of the project are:
Intellectual Merit:
This research addresses a critical need for formal techniques to
support the design of multiagent systems for distributed problem solving.
The increasing importance of sensor networks makes this a key application area
in which to deploy and promote MAS.
If successful, the proposed research would have an immediate impact by
substantially reducing the barriers to building large-scale sensor networks.
In addition, since many aspects of the proposed DT framework are not specific to
sensor networks or DSI, most results will have broader applicability.
Demonstrating that a principled and uniform approach can be used to design approximate,
satisficing MAS systems for real-world problems would have general intellectual value.
Broader Impact:
Successful completion of the project could have a substantial positive societal impact
by encouraging the development of sophisticated, large-scale sensor networks.
These networks can be used to enhance security and safety, as well as to make
computerized systems more autonomous and flexible.
The educational impacts from the proposed research will be quite significant, with
four graduate research assistants employed between UMass and SIUC.
At SIUC, this will provide the graduate students with exposure to high quality research
in a department that has somewhat limited opportunities at this level.
Results from the research will be disseminated as rapidly and widely as possible.
A project web site will be established, so that technical reports and other papers
can be made available as soon as possible.
Software that is to be developed for the project will be publicly released via the web site.
"Efficient Approximate Inference in Distributed Bayesian Networks for MAS-based Sensor Interpretation,"
N. Carver, to appear in
Proceedings of the 7th International Conference on Autonomous Agents and Multiagent Systems
(AAMAS08), 2008.
Available as:
PDF.
Decomposing Bayesian Network Representations of Distributed Sensor Interpretation Problems using
Weighted Average Conditional Mututal Information,
B. Haan, MS Thesis, Department of Computer Science, Southern Illinois University Carbondale, 2007.
Available as:
PDF.
"A New Framework for Inference in Distributed Bayesian Networks for Multi-Agent Sensor Interpretation,"
N. Carver,
Proceedings of the International Conference on Computers and Their Applications (CATA 2007),
March 2007.
Available as:
PDF.
"Communication Management Using Abstraction in Distributed Bayesian Networks,"
J. Shen and V. Lesser,
Proceedings of the Fifth International Joint Conference on Autonomous Agents and Multi-Agent Systems
(AAMAS06), 2006.
Available as:
PDF
"Agent Interaction in Distributed MDPs and its Implications on Complexity,"
J. Shen, R. Becker, and V. Lesser,
Proceedings of the Fifth International Joint Conference on Autonomous Agents and Multi-Agent Systems
(AAMAS06), 2006.
Available as:
PDF
Minimizing Communication Cost in an N-Agent Distributed Bayesian Network by using a Decentralized MDP,
J. Koren, MS Thesis, Department of Computer Science, Southern Illinois University Carbondale, 2005.
Available as:
gzipped Postscript or
PDF.
"Communication Management Using Abstraction in Distributed Bayesian Networks,"
J. Shen and V. Lesser,
Proceedings of the Fourth International Joint Conference on Autonomous Agents and Multiagent Systems
(AAMAS05), pp. 1115-1116 (extended abstract; presented as poster), 2005.
Available as:
PDF
"Analyzing the Efficiency of Strategies for MAS-based Sensor Interpretation and Diagnosis,"
N. Carver and R. Akavipat,
Proceedings of the Second International Joint Conference on Autonomous Agents and Multiagent Systems
(poster session), 2003.
Published (short) version available as:
gzipped Postscript or
PDF.
"Domain Monotonicity and the Performance of Local Solutions Strategies for CDPS-based
Distributed Sensor Interpretation and Distributed Diagnosis,"
N. Carver and V. Lesser,
International Journal of Autonomous Agents and Multi-Agent Systems,
Vol. 6, 35--76, 2003.
Available as:
gzipped Postscript or
PDF.
"Reasoning about Remote Data in CDPS with Distributed Bayesian Network,"
J. Shen, V. Lesser, and N. Carver,
Proceedings of Multi-Agent Systems and Applications -- ACAI,
July 2001.
Available as:
gzipped Postscript or
PDF.
"Nearly Monotonic Problems: A Key to Effective FA/C Distributed Sensor Interpretation?"
N. Carver, V. Lesser, and R. Whitehair,
Proceedings of AAAI-96, August 1996
(Copyright AAAI).
Available as:
gzipped Postscript.
Research Assistants:
Maxence Crossley (UMass)
Mark Garrett (SIUC)
Benjamin Haan (SIUC)
Jonathan Koren (SIUC, graduated)
Jiaying Shen (UMass)
Norman Carver's home page.
Victor Lesser's Multi-Agent Systems Laboratory page.